The Prediction Of Haze Concentration Based On Deep Learning

In recent years,the problem of haze pollution has become increasingly prominent,seriously affected people's work and life.So,the prediction of haze concentration has great significance.PM_2.5 is an important indicator of haze formation.The degree of haze pollution will become high as the rise of PM_2.5 concentration.Therefore,this paper studies the short-term prediction of PM_2.5 concentration.The specific researches of this paper are as follows:?1?Based on the previous research achievements,combined with the environment information of Xi'an,the paper use correlation analysis and stepwise regression analysis to determine the influencing factors of PM_2.5 concentration.The influencing factors of PM_2.5concentration are PM₁₀,CO,NO₂,SO₂,O₃,temperature,humidity,wind speed,air pressure and dew point.?2?By analysing the present study results of the PM_2.5 concentrationg prediction,the former has realized the fusion of the influence factors of PM_2.5 concentration on the spatial and temporal dimensions through chaining the Convolutional Neural Network?CNN?and the Long-term Memory Neural Network?LSTM?.However,this method will lose the time dimension characteristics of influencing factors of PM_2.5 concentration to some extent.Therefore,this paper uses Blending ensemble learning strategy to combine CNN and LSTM in parallel,and establish a PM_2.5 concentration prediction model based on CNN-LSTM ensemble learning.?3?This paper use the PM_2.5 concentration prediction model based on CNN-LSTM ensemble learning to predict the PM_2.5 concentration in Xi'an.For the comparative experiment,this paper establishs the model of PM_2.5 concentration prediction based on chaining CNN-LSTM,and use this model to predict the PM_2.5 concentration in Xi'an.Finally,through contrast experiments,it is known that the model established by conbining CNN and LSTM concurrently using Blending ensemble learning strategy is better than the prediction model of chaining CNN and LSTM in PM_2.5 concentration prediction.At the same time,the experimental results also show that the error of the prediction model increases,as the prediction duration increases.